There was an interesting anonymous comment by anon on yesterday’s Net Neutrality article:
if you have to treat all traffic on your network the same, you look closely at what traffic you put on the network.. settlement free peering is the next battle
I think I know what he means and I think he has a point worth exploring. Let me try to explain and perhaps readers with more extensive background in the nitty gritty of how the internet works will chime in and correct me.
Settlement free peering relationships between networks are simple ones. Two similar networks, each getting paid by their customers to deliver data, each agree to mutually hand off the traffic bound for the other over a connection they both promise to maintain and upgrade. Without the connection, they would need to pay a third party for transit somehow. But at the same time, settlement free peering relationships are immensely complicated because of the inherent ambiguity in the word ‘similar’, which must somehow cover everything from traffic levels to geographical territory, to company size and market share, and to different types of customers and traffic.
It is that last part that commenter ‘anon’ seems to be referring. The customers of last mile networks such as AT&T (NYSE:T, news, filings) and Comcast (NASDAQ:CMCSA, news, filings) are primarily eyeballs. The customers of over-the-top providers like Google and Akamai (NASDAQ:AKAM, news, filings) are primarily content sources. Others like glbc, Level 3 Communications (NYSE:LVLT, news, filings), and Cogent Communications (NASDAQ:CCOI, news, filings) lie at various spots between those two extremes. The traffic that originates from the two sides tends to be rather asymmetric, e.g. an end-user sends a few bits requesting a streaming movie and the content source sends a flood back. This asymmetry affects the economics of peering and hence peering relationships often contain restrictions on how balanced the traffic exchanged must be.
By forcing certain bandwidth hygiene on the last mile providers, network neutrality rules could shift that balance by elevating the importance of traffic ratios to one side. Google has all sorts of peering relationships with ISPs of all sizes, and Akamai’s entire network was built around peering with ISPs by putting servers in their closets. It’s not just them, the industry teams with peering relationships. The ISPs did it because it saved them money on transit and offered better performance for their customers. But if they must treat all traffic the same within their networks and traffic from content providers takes on a threatening hue, the economic case for maintaining those peering connections is weakened.
If an ISP must absorb all that traffic and isn’t allowed to treat it differently, why not just let that peering connection’s upgrades lag a bit and throttle traffic by inaction rather than action? For a tier-1 backbone with last mile networks, why peer at all in some cases? Why not try to make them accept paid peering or even buy transit? It’s not black and white, peering will still save both money in the end for most cases. But if the regulatory power balance between content and eyeballs shifts far enough, there will inevitably be some sort of echo in the peering/transit connections throughout the internet.
It could be a peaceful realignment, or it could escalate into something uglier as anon suggests. Any opinions out there?
If you haven't already, please take our Reader Survey! Just 3 questions to help us better understand who is reading Telecom Ramblings so we can serve you better!
Categories: Government Regulations · Internet Backbones · Internet Traffic
Join Our LinkedIn Group
Sorry, but you couldn’t be more wrong. It doesn’t matter which way traffic flows in a peering relationship. What matters is the amount of transit saved.
A tier 1 with consumer networks won’t accept peering from the likes of Google as it’s a Tier 1 and everyone pays it, except other Tier1’s so that point is moot.
Tier2’s do save on transit and for them it’s good practice to have peerings, regardless of the directionality of traffic.
If which way traffic flows is not relevant, than why would we need network neutrality? Peering may be a symmetrical event, but the relationships themselves are twisted by the endpoints. Otherwise why would peering agreements spend time specifying traffic ratios and such?
Perhaps a Tier1 and Google is the wrong example. Consider a Tier1 with lots of DSL/Cable customers and a Tier1 with a majority of Content customers (e.g. Cogent). There has always been friction there, but isn’t it likely to get worse?
There has been much argument about Google’s peering relationships and their effect on its cost structure. If no Tier-1 providers peered with Google, they wouldn’t find much savings – some clearly do. Granted some of it is likely paid peering of some sort, but my point is that if the economics change, those relationships will also change one way or another.
Net Neutrality has nothing to do with peering and transit.
There are articles on the Net called the Folly of peering ratios. Just because the people from the voice world work with symmetric traffic balances doesn’t mean it’s right.
Google may not peer with Tier1’s. It doesn’t matter. Most of the world lives in a world where they need to pay transit for part of their traffic-bill. So they are TierX.
Rob, I think you’re viewing the matter in linear, chart-like terms, where all of the consumers are on the right of the continent, and all of the content providers on the left. This may be a convenient prop for chalk-talk an power-point purposes, but not how real life works. Interspersed throughout the topology you have both, so a great deal of balancing is taking place on the whole. This is less through design than by sheer happenstance.
Maybe the government should invest and run peering points to assure whatever it is they think the market can’t provide … IP access costs at zero – a game changer …
Translated: the government would fail. They should try first before screwing with the private sector.
thanks Rob for picking up on this discussion. i think your description of the context was right on. a few clarifications: traffic ratio’s are important — always looked at by carriers in evaluating a potential peer (if all of your traffic is “outbound” like a hosting provider, hard to get someone to “peer” with you. conversely, the large isp’s (e.g., cable mso’s) have a ton of unused outbound (“to the cloud”) capacity. Sprint, for one, is rumored to closely watch ratio’s and will enforce protections against imbalanced peers.
The larger point is that there are the true tier 1 carriers that have a ton of inbound/outbound traffic, global routes, etc. and have no issue here (e.g., att will continue to peeri with DT and VZ). But what of cogent? and the so called “transit networks” like GX, Above, etc that basically connect networks together (fewer eyeballs). Seems unlikely that the Tier 1 guys with their HUNDREDS of MILLIONS OF EYEBALLS will let someone reach them for free… How long will they permit “over the toppers” to buy $2 bandwidth from a lesser network and then take all the traffic in settlement free peering – especially with the new “neutrality” obligations… the only thing for sure is that every action has a reaction… tier 1’s have a Hundred Billion ++ in legacy capital to honor…
Crowe has been speaking up lately on how peering will have to change/next thing to be looked at.
Are these the catalysts?
If preserved peering relationships lapse as the AT&T & VZ merger deal conditions run off while peering gets reset to balance the network neutrality issue – which networks rely most on peering?
I would guess Cogent. Who else is getting more out of the kiddy than what they put in?
“How long will they permit “over the toppers” to buy $2 bandwidth from a lesser network and then take all the traffic in settlement free peering”
That depends on how long they want “over the toppers” deploying their own fibre and servers deep in their networks. The net result in charging Limelight, Google and Akamai for peering is to throw 5x more traffic on the edges of their backbone network, which will promptly fall apart.
Of course the direction of traffic flows matter. They are absolutely central. And that’s because of a simple term “hot potato routing”. When one network sends traffic to another network they find a point of interconnect that is closest to the origination of the traffic. This means they limit to a minimum the distance that traffic is carried over their own network.
So image two peers. One with content in LA. The other with an eyeball in NYC. They both build a network between those two cities and peer in both places.
A request for content from the eyeball originates in NYC. The eyeball network then hot-potato routes it onto the content owners network in NYC. The content owners network carriers it all the way to LA. And conversely in LA the content owner hot potato routes it onto the eyeball owners network. The content is then carried from LA over the eyeball owners network back to NYC.
A few bytes in one direction. Potentially gigabytes in the other direction.
The content owner had to build a very skinny network in order to peer. The eyeball owner had to build a very fat network in order to peer.
That’s a simplification for sure but designed to demonstrate why traffic balance is vital in peering relationships.
An interesting question is “How much value is an eyeball network to its customers if they cannot get to content quickly?” It seems to me that the value of a high speed internet connection to an end user is the ability to access content. If my high speed wireless/cable/DSL is slow getting to Google, Yahoo, or even more so (from a business perspective) to hosted business apps – what is my money buying me?